This systematic review and meta-analysis were performed in compliance with the Cochrane Handbook for Systematic Reviews of Interventions (19) and was reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (20). The review protocol was prospectively registered in International Platform of Registered Systematic Review and Meta-analysis Protocols (INPLASY) under number INPLASY202170032. The PRISMA 2020 checklist is available in the Supplementary Table 1.

We considered studies to be eligible if they met the following criteria: (i) Types of studies: RCT published in peer-reviewed medical journals; (ii) Types of participants: Adult patients (aged ≥18) with spontaneous ICH and are treatable within 24 h of symptom onset; (iii) Types of interventions: TXA at any dose versus placebo; (iv) Types of outcome measures: The outcome measurements were HE (defined as hematoma growth >33% and/or >6 ml), 3-month poor functional outcome (PFO; defined as modified Rankin Scale score 4–6), 3-month mortality, and major thromboembolic events (MTE; according to the definition provided in each study).

We conducted a comprehensive literature search on Cochrane Central Register of Controlled Trials (CENTRAL), ClinicalTrials.gov, Embase, PubMed, and the World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) from inception until July 3, 2021 with no language restriction. RCTs investigating the effects of TXA in adult patients with acute spontaneous ICH were included. Controlled vocabulary (i.e., MeSH and Emtree) and keywords were used. Search terms included TXA, ICH, RCT, and their variants. The complete search strategy is available in the Supplementary Table 2. We manually screened the reference lists of eligible trials and previous relevant reviews for additional studies.

After records were imported into the Zotero reference management software (www.zotero.org), duplicate records were manually removed. Two reviewers independently screened the titles and abstracts for relevance, and labeled records as probably included and excluded in duplicates. If records are deemed potentially relevant by either reviewer, the full-text articles were retrieved to assess its eligibility. Disagreements were resolved by discussion, and by a third-party adjudication if required.

Two reviewers independently extracted data and in duplicate using a standardized form. We extracted the following information from included trials: (i) trial characteristics: study title, fist author name, year of publication, country of origin, study design, and number of participants; (ii) patient characteristics: CT markers of HE, age, sex, and baseline National Institutes of Health Stroke Scale (NIHSS) score; (iii) intervention characteristics: treatment administration time and TXA dose; and (iv) data on outcomes of interest, etc.

Two reviewers have assessed the risk of bias of each trial independently and in duplicates using the Cochrane Collaboration's tool based on the recommendations of the Cochrane Handbook for Systematic Reviews of Intervention (21). The following items for risk of bias were examined: random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete data outcome (attrition bias), selective reporting (reporting bias), and, other biases (such as stopping early and funding source). The risk of bias was determined as “high risk,” “unclear risk,” or “low risk.” Disagreements were resolved by discussion, and third-party adjudication if required.

Two reviewers assessed the overall certainty of evidence for each outcome using the Grading Recommendations Assessment, Development and Evaluation (GRADE) system (22). The items assessed included risk of bias, inconsistency, indirectness, imprecision, and publication bias. The evidence quality was classified as very low, low, moderate, and high. Disagreements were resolved by discussion, and third-party adjudication if required.

We calculated odds ratios (ORs) and their corresponding 95% CIs to measure the effect size while comparing TXA vs. control among patients with acute spontaneous ICH. Meta-analyses were performed using DerSimonian and also Laird random-effects models accounting for clinical heterogeneity (23). A p-value of < 0.05 was considered statistically significant. The heterogeneity between trials was assessed using the Cochran Q test (with p < 0.1 indicating significance) and was quantified by the I² statistic (for which a value of 50% or greater was considered to represent significant heterogeneity) (24). Publication bias across individual trials was graphically evaluated using a funnel plot and also with the Egger's test at a significance level of p < 0.05 (25). We performed one post hoc sensitivity analysis, restricting to trials with a low risk of bias, to test the robustness of our findings. We also conducted subgroup analysis according to different subjects (high-risk population or standard-risk population) and the time from onset to randomization ( ≤ 4.5 or >4.5 h). High-risk population was defined as patients with CT markers of HE in patients with ICH. Standard-risk population was defined as general ICH population, without further screening of CT markers of HE. Specific data of the intervention and control groups were extracted from publications. If the original data were not available, the adjusted ORs, as the indicators of the outcomes, were alternatively extracted. All statistical analyses were conducted with Comprehensive Meta-Analysis software (version 3) and Stata software (version 15).

The initial search yielded 702 records. We excluded 275 duplicates and a further 427 records after title and abstract screening. Then, 20 potentially eligible articles were retained for full-text evaluation. Finally, seven trial reports (26–32) representing 5 RCTs were included according to the inclusion criteria. Figure 1 illustrates the study selection process.

Baseline characteristics of the included seven trial reports are summarized in Table 1. Of the two additional trial reports, one was a post hoc analysis (31), and the other was a pre-specified subgroup analysis (32) of a previously published trial (30). The 7 trial reports were published from 2014 to 2021, with sample sizes ranging from 24 to 2,325 subjects, and a total of 2,650 subjects. Three trials were multicenter studies (27, 28, 30), while two were conducted at a single center (26, 29). The subjects were mainly men and the mean age of subjects ranged from 53 to 73 years. The baseline NIHSS scores ranged from 12 to 18. Two trials (27, 28) enrolled acute spontaneous ICH patients susceptible to HE based on imaging assessment; however, the majority of patients included were general acute spontaneous ICH patients. The TXA administration time was varied among studies: within 4.5 h in one trial (28), within 8 h in three trials (26, 27, 30), and, up to 24 h in one trial (29). The dosage of TXA was consistent across included trials with the most common regimen being a loading dose of 1 g, followed by a maintenance dose of 1 g over 8 h. Overall, three trials (27, 28, 30) were categorized as being at low risk of bias and 2 (26, 29), as being unclear. Details of the risk of bias are presented in Figure 2 and Supplementary Table 3.

Table 2 summarizes the overview of the association between TXA and various clinical outcomes. GRADE summary of findings is available in the Table 2 and Supplementary Table 4.

Pooled analysis found that TXA administration may reduce HE on subsequent neuroimaging (OR 0.825; 95 % CI 0.692–0.984; p = 0.033; I² = 0 % and chi-square p = 0.732; Figure 3A). The GRADE certainty of the evidence was moderate. Sensitivity analysis showed similar results when limited to trials with low risk of bias (OR 0.828; 95% CI 0.693–0.988; p = 0.036; I² = 0 % and chi-square p = 0.935; Supplementary Appendix 1). Subgroup analyses indicated that TXA reduced the risk of HE in the high-risk population (OR 0.646; 95% CI 0.503–0.829; p = 0.001; I² = 0 % and chi-square p = 0.886; Figure 4B) and patients who were treatable within 4.5 h of symptom onset (OR 0.823; 95% CI 0.690–0.980; p = 0.029; I² = 0% and chi-square p = 0.690; Figure 5A), but not in the standard-risk population (OR 0.834; 95% CI 0.690–1.007; p = 0.059; I² = 0 % and chi-square p = 0.388; Figure 4A), and patients who were treatable over 4.5 h of symptom onset (OR 1.026; 95% CI 0.795–1.324; p = 0.844; I² = 0 % and chi-square p = 0.399; Figure 5B).

However, TXA probably had no effect on a 3-month PFO (OR 0.991; 95 % CI 0.849–1.158; p = 0.914; I² = 0 % and chi-square p = 0.871; GRADE: high certainty; Figure 3B), 3-month mortality (OR 1.020; 95 % CI 0.843–1.234; p = 0.834; I² = 0 % and chi-square p = 0.623; GRADE: high certainty; Figure 3C), and MTE (OR 1.092; 95% CI 0.721–1.655; p = 0.678; I² = 0 % and chi-square p = 0.848; GRADE: high certainty; Figure 3D). To confirm the robustness of our findings, we conducted sensitivity analyses using data from trials categorized as low risk of bias for various clinical outcomes. Sensitivity analysis for the rate of each endpoint showed that the overall effect of TXA therapy was consistent with the overall estimates from all studies (Supplementary Appendix 1). There was no significant difference in the 3-month PFO, 3-month mortality, and MTE in the high-risk population vs. standard-risk population subgroups (Supplementary Appendix 2).

For the safety and efficacy analyses on different endpoints, funnel plot and the Egger's test revealed no evidence of asymmetry (Supplementary Figure 1). However, the results from such analyses should be treated with considerable caution based on a limited number of trials.

Our meta-analysis has comprehensively and systematically reviewed the current available literature that compared TXA with placebo for treating acute spontaneous ICH, and we obtained three major findings. Firstly, in patients with spontaneous ICH, TXA may decrease HE on subsequent imaging. However, TXA probably had no effect on PFO and on mortality at 3 months. The use of TXA was safe without increasing thromboembolic complications. Secondly, further analyses restricted to patients with spontaneous ICH susceptible to HE based on imaging assessment have showed consistent results, but this protective effect was not found in the standard-risk population. CT markers of HE may be a useful determinant of benefit from TXA administration in acute spontaneous ICH patients. Thirdly, early treatment (within 4.5 h after stroke onset) seemed to show more benefits in preventing HE, but this protective effect was not observed in the patients who were treatable over 4.5 h.

Two previous systematic reviews on the similar topic have been published (33, 34). One of them evaluated the treatment of acute spontaneous ICH with different hemostatic therapies (including TXA) and the results showed that there was no evidence of either benefit or harm from TXA for people with ICH (33). The other one evaluated the effect of TXA on acute spontaneous ICH and found that TXA could reduce HE in ICH (34). In line with the second study (34), our meta-analysis also found that TXA may decrease HE on subsequent imaging. Besides, we found that the decreased risk was observed in patients who were treatable within 4.5 h and in high-risk population, but not in those who were treatable over 4.5 h and in standard-risk population. In summary, our meta-analysis further confirms that TXA is an effective and safe drug for the treatment of acute spontaneous ICH and that CT markers of HE, as well as early time windows, are useful clinical determinants of TXA administration in acute spontaneous ICH patients. However, some differences also should be noted. First, previous meta-analyses included <4 RCTs. Our meta-analysis identified another three recent studies and further reinforced earlier results of previous meta-analyses. Second, we pooled RCTs data with a random-effects model accounting for clinical heterogeneity to ensure a more conservative estimate of the efficacy and safety of TXA for the treatment of acute spontaneous ICH. Third, we evaluated the certainty of evidence using GRADE approach to facilitate clinical decision-making.

This systematic review and meta-analysis have several strengths including a pre-registered protocol, a comprehensive literature search, a duplicate and independent screening and data extraction, and GRADE assessment of certainty of evidence. However, certain limitations of this meta-analysis need to be acknowledged. First, subgroup analysis for each CT markers of HE was not performed because of the lack of data. Thus, the difference in efficacy for patients with different CT markers of HE was not determined. Second, the inclusion of the post hoc RCT would lead to a declined quality of subgroup analysis that we conducted.

The study limitations mentioned above provided several considerations for future studies. In particular, the difference in efficacy for patients with different CT markers of HE should be determined. Subgroup analyses according to time from onset to randomization have found that early treatment (within 4.5 h) seemed to show more benefit. The proposed time window could also be used for hemostatic therapy in future research.